The group wants its staff-facing agent to answer "what is our post-operative discharge policy for orthopaedic patients" from approved documents rather than from the model's own knowledge. Which configuration concept addresses this requirement?
Select an answer to reveal the explanation.
Short Explanation
Telling a model to be accurate is a suggestion. Handing it the actual document is an answer. Grounding is the difference between an agent that sounds confident and an agent that is quoting your own approved post-op instructions back to you.
Full Explanation
Grounding is the mechanism that supplies an agent with trusted, org-specific content at run time, whether that comes from CRM records, knowledge articles, a Flow or Apex action, or a Data 360 retriever over indexed documents. It is the only one of these choices that changes where the answer comes from. Guardrail instructions about accuracy are the classic trap: they feel like the right lever and they do shape behaviour, but an instruction to be accurate cannot supply facts the agent does not have, so an ungrounded agent will simply produce a confident invention in a more careful tone. Limiting response length constrains verbosity, not provenance; a short ungrounded answer is still ungrounded. Switching agent type changes who the agent serves, which standard topics and actions are available, and the security model, but an Employee agent is already the right type for internal staff here and the switch would not by itself connect the discharge documents. The exam-relevant nuance: for unstructured PDFs such as discharge instructions, grounding runs through a Data Library with chunking and indexing and a retriever, whereas structured policy stored on records can be grounded directly through a record-based action.